Machine Learning-Powered
Search Tool
Driving discovery and seller success through barcode and image-based workflows at Alias (GOAT Group)
Role
Lead Product Designer
Timeline
Jun 2021 – Mar 2024
Company
Alias by GOAT Group
Designed a seamless barcode and image search experience that improved adoption, accuracy, and seller confidence.
Challenge
Sellers struggled to discover items and our text-only search required prior knowledge, limiting adoption and engagement.
Strategy
- Designed a two-phase ML search workflow: MVP (photo search) and Phase 2 scaling (barcode-first + image).
- Integrated ML for accurate search results and used seller survey insights to inform design decisions.
Impact
- Improved scan accuracy (UPC ~90%, image ~60%).
- Increased adoption and satisfaction (Visual Search 5% → 13%; app rating 4.0 → 4.4).
- Built a scalable foundation for future ML improvements (ongoing data capture + feedback loop).
Phase 1: MVP & Validation
Key Action #1
Partnered with Data Science to optimize predictions
Worked with data science to surface the top 4 predictions. Photo capture and upload became primary search methods, with images displayed alongside results for easy comparison.
Key Action #2
Conducted field testing to validate model accuracy
Tested photos in-store across angles and lighting, confirming that leveraging existing sneaker templates improved ML prediction reliability.
Key Action #3
Proposed barcode scanning to enhance accuracy
Assessed ML constraints and suggested adding barcode scanning to improve search accuracy. Leadership simplified the MVP, so we guided users to try UPC input through instructional copy instead.
Key Action #3
Introduced a feedback loop
Recommended a “Rate the App” flow to capture user insights and provide data for ongoing model improvement.
Final MVP Design
Seamless photo upload and capture with dynamic results and a feedback loop.
MVP Outcomes
- Widely adopted; app rating ↑ 4.0 to 4.4
- Set foundation for Phase 2
Phase 2: Scale and Refine
The feature was picked up again to scale adoption and tackle key pain points identified by sellers.
Research: Seller Survey Insights
- 80% rated Visual Search 4/5 or 5/5
- 24% secondary use as price checking
- Main pain point = Inaccurate scans
Key Action #1
Shifted to barcode-first workflow → improved accuracy and consistency
Survey insights showed scan accuracy was the main pain point. Prioritizing barcode input streamlined user flows and improved reliability across sneaker searches.
Key Action #2
Defined interaction structure
Organized barcode scan + photo capture tabs, and results page for clarity. Simplified design and added scan animation to reduce confusion and guide user behavior.
Key Action #3
Iterated and aligned cross-functionally
Incorporated stakeholder feedback and collaborated with engineering, brand, and copy teams to create a polished, consistent experience across barcode and image capture flows.
Final Design
Optimized search experience with barcode-first input, clear results, and polished interactions.
Phase 2 Outcomes
- Visual Search = 13% of all searches (up from 5%)
- UPC scan success ~90%; image search ~60%
Reflections
Designing for AI/ML requires brutal clarity.
Translating model limits and confidence thresholds into simple, guided UX was the only way to improve accuracy without overwhelming users.
Feedback loops are non-negotiable.
AI features don’t improve passively. Intentional data capture and quick iteration are what allowed the product to keep getting smarter over time.
Other Projects
Search and Discovery Redesign
→
Fulfillment Internal Tool
Fulfillment Internal Tool
→
w
d
CONTACT
workudianne@gmail.com
SOCIAL
Machine Learning-Powered Search Tool
Driving discovery and seller success through barcode and image-based workflows at Alias (GOAT Group)
Role
Lead Product Designer
Timeline
Jun 2021 – Mar 2024
Company
Alias by GOAT Group
Designed a seamless barcode and image search experience that improved adoption, accuracy, and seller confidence.
Challenge
Sellers struggled to discover items and our text-only search required prior knowledge, limiting adoption and engagement.
Strategy
- Designed a two-phase ML search workflow: MVP (photo search) and Phase 2 scaling (barcode-first + image).
- Integrated ML for accurate search results and used seller survey insights to inform design decisions.
Impact
- Improved scan accuracy (UPC ~90%, image ~60%).
- Increased adoption and satisfaction (Visual Search 5% → 13%; app rating 4.0 → 4.4).
- Built a scalable foundation for future ML improvements (ongoing data capture + feedback loop).
Phase 1: MVP & Validation
Key Action #1
Partnered with Data Science to optimize predictions
Worked with data science to surface the top 4 predictions. Photo capture and upload became primary search methods, with images displayed alongside results for easy comparison.
Key Action #2
Conducted field testing to validate model accuracy
Tested photos in-store across angles and lighting, confirming that leveraging existing sneaker templates improved ML prediction reliability.
Key Action #3
Proposed barcode scanning to enhance accuracy
Assessed ML constraints and suggested adding barcode scanning to improve search accuracy. Leadership simplified the MVP, so we guided users to try UPC input through instructional copy instead.
Key Action #4
Introduced a feedback loop
Recommended a “Rate the App” flow to capture user insights and provide data for ongoing model improvement.
Final MVP Design
Seamless photo upload and capture with dynamic results and a feedback loop.
MVP Outcomes
- Widely adopted; app rating ↑ 4.0 to 4.4
- Set foundation for Phase 2
Phase 2: Scale and Refine
The feature was picked up again to scale adoption and tackle key pain points identified by sellers.
Seller Survey Insights
- 80% rated Visual Search 4/5 or 5/5
- 24% secondary use as price checking
- Main pain point = Inaccurate scans
Key Action #1
Shifted to barcode-first workflow → improved accuracy and consistency
Survey insights showed scan accuracy was the main pain point. Prioritizing barcode input streamlined user flows and improved reliability across sneaker searches.
Key Action #2
Defined interaction structure
Organized barcode scan + photo capture tabs, and results page for clarity. Simplified design and added scan animation to reduce confusion and guide user behavior.
Key Action #3
Iterated and aligned cross-functionally
Incorporated stakeholder feedback and collaborated with engineering, brand, and copy teams to create a polished, consistent experience across barcode and image capture flows.
Final Design
Optimized search experience with barcode-first input, clear results, and polished interactions.
Phase 2 Outcomes
- Visual Search = 13% of all searches (up from 5%)
- UPC scan success ~90%; image search ~60%
Reflections
Designing for AI/ML requires brutal clarity.
Translating model limits and confidence thresholds into simple, guided UX was the only way to improve accuracy without overwhelming users.
Feedback loops are non-negotiable.
AI features don’t improve passively. Intentional data capture and quick iteration are what allowed the product to keep getting smarter over time.
Other Projects
Search and Discovery Redesign
→
Fulfillment Internal Tool
Fulfillment Internal Tool
→
w
d
CONTACT
workudianne@gmail.com
SOCIAL
Machine Learning-Powered Search Tool
Driving discovery and seller success through barcode and image-based workflows at Alias (GOAT Group)
Role
Lead Product Designer
Timeline
Jun 2021 – Mar 2024
Company
Alias by GOAT Group
Designed a seamless barcode and image search experience that improved adoption, accuracy, and seller confidence.
Challenge
Sellers struggled to discover items and our text-only search required prior knowledge, limiting adoption and engagement.
Strategy
- Designed a two-phase ML search workflow: MVP (photo search) and Phase 2 scaling (barcode-first + image).
- Integrated ML for accurate search results and used seller survey insights to inform design decisions.
Impact
- Improved scan accuracy (UPC ~90%, image ~60%).
- Increased adoption and satisfaction (Visual Search 5% → 13%; app rating 4.0 → 4.4).
- Built a scalable foundation for future ML improvements (ongoing data capture + feedback loop).
Phase 1: MVP & Validation
Key Action #1
Partnered with Data Science to optimize predictions
Worked with data science to surface the top 4 predictions. Photo capture and upload became primary search methods, with images displayed alongside results for easy comparison.
Key Action #2
Conducted field testing to validate model accuracy
Tested photos in-store across angles and lighting, confirming that leveraging existing sneaker templates improved ML prediction reliability.
Key Action #3
Proposed barcode scanning to enhance accuracy
Assessed ML constraints and suggested adding barcode scanning to improve search accuracy. Leadership simplified the MVP, so we guided users to try UPC input through instructional copy instead.
Key Action #4
Introduced a feedback loop
Recommended a “Rate the App” flow to capture user insights and provide data for ongoing model improvement.
Final MVP Design
Seamless photo upload and capture with dynamic results and a feedback loop.
MVP Outcomes
- Widely adopted; app rating ↑ 4.0 to 4.4
- Set foundation for Phase 2
Phase 2: Scale and Refine
The feature was picked up again to scale adoption and tackle key pain points identified by sellers.
Seller Survey Insights
- 80% rated Visual Search 4/5 or 5/5
- 24% secondary use as price checking
- Main pain point = Inaccurate scans
Key Action #1
Shifted to barcode-first workflow → improved accuracy and consistency
Survey insights showed scan accuracy was the main pain point. Prioritizing barcode input streamlined user flows and improved reliability across sneaker searches.
Key Action #2
Defined interaction structure
Organized barcode scan + photo capture tabs, and results page for clarity. Simplified design and added scan animation to reduce confusion and guide user behavior.
Key Action #3
Iterated and aligned cross-functionally
Incorporated stakeholder feedback and collaborated with engineering, brand, and copy teams to create a polished, consistent experience across barcode and image capture flows.
Final Design
Optimized search experience with barcode-first input, clear results, and polished interactions.
Phase 2 Outcomes
- Visual Search = 13% of all searches (up from 5%)
- UPC scan success ~90%; image search ~60%
Reflections
Designing for AI/ML requires brutal clarity.
Translating model limits and confidence thresholds into simple, guided UX was the only way to improve accuracy without overwhelming users.
Feedback loops are non-negotiable.
AI features don’t improve passively. Intentional data capture and quick iteration are what allowed the product to keep getting smarter over time.
Other Projects
Search and Discovery Redesign
→
Fulfillment Internal Tool
Fulfillment Internal Tool
→
w
d
CONTACT
workudianne@gmail.com
SOCIAL